This paper investigates the equilibrium exchange rates of three Southeastern European countries, namely Bulgaria, Croatia and Romania, of two CIS economies, namely Russia and Ukraine, and of Turkey. A systematic approach in terms of different time horizons at which the equilibrium exchange rate is assessed is conducted, combined with a careful analysis of country-specific factors. The deviation from absolute purchasing power parity (PPP) and from the real exchange rate, which is given by relative productivity levels, is investigated. For Russia, a first look is taken at the Dutch disease phenomenon as a possible driving force behind equilibrium exchange rates. As a next step, a Behavioral Equilibrium Exchange Rate (BEER) model including productivity and net foreign assets is estimated using both time series and panel techniques. Control variables such as openness, public debt and public expenditures are also used to check for the robustness of the results. In a final stage, total real misalignment bands are computed for the countries under study.
1 Introduction
The prospect of joining the EU and the actual accession of eight countries from Central and Eastern Europe to the European Union in May 2004 have drawn much attention to these countries equilibrium exchange rates. By contrast, equilibrium exchange rates of countries in Southeastern Europe and of the CIS have been less in focus and only a few papers have investigated this issue. A considerable number of the papers on this topic deal with these coun- tries in a panel context, which may be insufficient for accounting for country- specific features.2Only very few studies analyze Bulgaria, Croatia, Romania and Russia.3
In this paper, we seek to fill this gap by investigating the equilibrium exchange rates of three Southeastern European countries, namely Bulgaria, Croatia and Romania, of two CIS economies, namely Russia and Ukraine, and of Turkey. These countries are of interest because Bulgaria, Romania and prob- ably Croatia will join the EU in the foreseeable future, and Russia, Ukraine and Turkey are of utmost economic interest for the EU in the (South)eastern part of Europe. For this purpose we have proposed a systematic approach in terms of different time horizons at which the equilibrium exchange rate is assessed, com- bined with a careful analysis of country-specific factors. Questions related to equilibrium exchange rates in these countries as compared to those in the new Member States in Central and Eastern Europe can either be raised differently or are sometimes truly different. For instance, the issue of current account and foreign debt sustainability is of utmost importance for Croatia. Also, the real exchange rate in Russia may be driven not only by traditional channels but also by the Dutch disease phenomenon.
1 Oesterreichische Nationalbank (OeNB), Foreign Research Division, [email protected], and MODEM, University of Paris X-Nanterre, [email protected], and William Davidson Institute.
Many thanks go to Ales´ Delakorda, Anna Dorbec, Jarko Fidrmuc, Nienke Oomes, Thomas Reininger and Doris Ritzberger- Gru‹nwald for very useful comments and suggestions, to Zolta«n Walko for help in constructing some of the time series used in the paper, and to Alexandra Edwards and Irene Mu‹hldorf for excellent language advice. The opinions expressed in the paper are those of the author and do not necessarily represent the views of the Oesterreichische Nationalbank or the European System of Central Banks (ESCB).
2 E.g. Halpern and Wyplosz (1997, 2001), Krajnya«k and Zettelmeyer (1998), Begg et al. (1999), De Broeck and Sl¿k (2001), Dobrinsky (2003) and Fischer (2004).
3 Chobanov and Sorsa (2004) analyze Bulgaria. Stapafora and Stavlev (2003), Sosunov and Zamulin (2004) and Rautava (2004) study the case of Russia. Crespo-Cuaresma et al. (2004) apply the monetary model to Bulgaria, Croatia, Romania and Russia.
Bala«zs E«gert1
First, we take a look at the deviation from absolute PPP. Subsequently, we investigate whether the real exchange rates in levels correspond to the under- lying productivity levels. In a next step, the factors of real exchange rate move- ments are studied. First, the simple Balassa-Samuelson effect and the Dutch disease are put under the microscope. In a next step, the stock-flow approach is used to widen the horizon. Both time series and panel data are used to study deviations from the equilibrium exchange rate.
The remainder of the paper is structured as follows: Section 2 briefly addresses the deviation from absolute PPP. Section 3 investigates the relation- ship between the level real exchange rates and relative productivities. Section 4 analyzes factors behind a possible trend appreciation of the currencies. The Balassa-Samuelson effect is studied in great detail, and a first look is taken at the Dutch disease in Russia. A Behavioral Equilibrium Exchange Rate (BEER) model including productivity and net foreign assets is estimated using both time series and panel techniques. Control variables such as openness, public debt and public expenditures are also used to check for the robustness of our results.
Finally, total real misalignment bands are computed for the countries under study. Section 5 presents some concluding remarks.
2 Undervaluation in Terms of Purchasing Power Parity
In the paper, we follow a bottom-up approach in that we start looking at approaches to the equilibrium exchange rate which are assumed to hold in the long run. We then move forward systematically toward shorter time hori- zons.
Let us now begin with the concept of PPP, which can be thought of as a very long-term approach for countries in the catching-up process. It is a well-under- stood fact that PPP is a poor tool, even in the long run, for measuring equili- brium exchange rates for transitional and developing economies because their currencies are undervalued in terms of PPP. According to PPP, the exchange rate given by the ratio of domestic and foreign absolute price levels should be equal to the nominal exchange rate which can be observed on the foreign exchange market. In other words, the real exchange rate, which is given as
E=ðP =PÞ ¼EP=P, should equal 1. With the exchange rate being defined as domestic currency units expressed in terms of one unit of foreign currency,4 a real exchange rate higher than 1 implies undervaluation, which can be clearly observed vis-a‘-vis the euro for all countries under study (see table 1). The larg- est undervaluation has been found in Ukraine, whereas the Croatian currency appears to be the least undervalued one among the countries. There are evident signs of a steady decrease in undervaluation for Bulgaria, Romania and perhaps for Russia. By contrast, the undervaluation appears pretty stable for Croatia and Turkey, and it fluctuates strongly for Ukraine.5
4 In the rest of the paper, an increase/decrease in the (real) exchange rate implies a depreciation/appreciation.
5 For Russia and Ukraine, some of the fluctuations may be due to changes in the euro/U.S. dollar exchange rate.
3 The Role of Productivity:
A Cross-Sectional Perspective
According to the traditional Balassa-Samuelson argument, the less developed country is usually less productive in producing tradable goods. The price level in the open sector is given by the PPP condition. At the same time, the level of productivity in the open sector, which is usually lower in the less developed country, determines the price level in the closed sector through intersectoral wage linkages. Hence, the price level in the sheltered sector, and subsequently the overall price level, will be below that prevailing in the more developed country. As a result, the observed nominal exchange rate given by PPP in the open sector appears to be weaker (higher) than the exchange rate given by PPP.
Notice, however, that this undervaluation in PPP terms is an equilibrium undervaluation if it reflects a difference between productivity levels. By con- trast, it may be the case that the price level does not fully reflect productivity levels. If prices are higher than what productivity levels would predict, the exchange rate can be viewed as overvalued in terms of productivity levels (although still undervalued in PPP terms). If prices are lower than what produc- tivity levels would predict, the currency can be thought of as undervalued (not only in PPP terms). This is depicted in chart 1.
We now set out to analyze whether the exchange rates of the countries under consideration were undervalued or overvalued in terms of productivity levels. Put differently, we are interested in whether a given country is at point A, A or A in chart 2. Such an analysis is best conducted using cross-sectional data. In such a framework, the real exchange rate in levels or the relative price level of the home country vis-a‘-vis a benchmark economy (the reciprocal of the real exchange rate in levels) is regressed on the relative productivity level of
Deviation from Absolute Purchasing Power Parity vis-à-vis the Euro
8 7 6 5 4 3 2 1
Chart 1
Source: Author,s calculations based on data obtained from the wiiws annual database. The data for Turkey were obtained from NewCronos/Eurostat.
Note: The charts are obtained as EP*/P, where E is the actual nominal exchange rate, and P and P* are the absolute domestic and foreign prices.
19931996 1999
Bulgaria Croatia Romania Russia Ukraine
20022003
Turkey Real exchange rate level
the home country to that in the foreign benchmark. In practice, however, GDP per capita or GDP per employment expressed in PPP terms, which is a broad proxy for productivity, is employed based on data (un)availability.
A number of studies have already investigated this relationship extensively.
We make use of the regression results reported in these studies. We have selected all the equations which use the EU-15 as the foreign benchmark.
The five retained equations from three papers, namely C´ iha«k and Holub (2003), Coudert and Couharde (2003) and Maeso-Fernandez et al. (2004) are reported in table 1. The fitted values of the real exchange rates in level or relative price levels of the countries under study obtained from these equa- tions are then compared to the actual real exchange rates or relative price levels for each country against the EU-15.
The three papers offer an interesting combination of country coverage.
Coudert and Couharde (2003) include 120 developing and emerging econo- mies, whose GDP per capita expressed using the purchasing power standard did not exceed the corresponding figure of the euro area. The sample also included all transition economies with a few exceptions. By contrast, the sample used in Maeso-Fernandez et al. (2004) is composed of 25 industrialized OECD countries, excluding all transition economies.6C´ iha«k and Holub (2003) keep to
Chart 2
Source: Égert, Halpern and MacDonald (2005), Égert (2003).
undervalued both in terms of PPP and productivity levels Q: level of the real exchange rate
A'' A
B
overvalued in terms of productivity C
levels, but still undervalued in terms of absolute PPP
A'
quick appreciation toward equilibrium
PPP zone corresponding to prevailing productivity and price levels B-S effect and tradable price-based trend appreciation toward PPP target because of
speedy economic catching-up
Trend Appreciation of the Equilibrium Real Exchange Rate
equilibrium real exchange rate
PPP zone 1+m
1 1m
time FEER, BEER, medium-term NATREX: 38 years
Long-run NATREX: 515 years
B-S effect and tradable price-based real appreciation: 1530 years PPP: quickly catching-up countries 30100 years, much shorter
otherwise undervalued in terms of
PPP, but fairly valued in terms of productivity
levels
D
6 The panel includes the EU-15 (without Luxembourg), Australia, Canada, New Zealand, the U.S.A., Norway, Iceland, Korea, Mexico and Turkey. OECD countries such as the Czech Republic, Hungary, Poland and Slovakia are excluded.
the golden mean in that a number of EU-15 countries and transition economies are used together for their estimations.
These observations have interesting implications. First, the regression based on a large number of developing and emerging countries can be viewed as reflecting how the real exchange rate and per capita GDP may be linked, on average, in emerging and developing economies. Second, using a narrow sample of industrialized countries offers some perspectives regarding what this relationship looks like for higher GDP per capita levels. For the countries under study, such a relationship could be thought of as applying in the longer run (because the developing and emerging economies are expected to catch up with the industrialized economies in the long run). Third, taking a group of European transition and developed EU economies may tackle some hetero- geneity problems in Coudert and Couharde (2003) and, at the same time, helps anticipate the long-term behavior given by the regression results in Maeso- Fernandez et al. (2004).
Chart 3 reports under- and overvaluations in terms of productivity levels for the period of 1991 to 2003.7For Bulgaria and Romania, the real exchange rates seem to have been undervalued at the beginning of the 1990s. This is something which is labeled initial undervaluation by Halpern and Wyplosz (1997), and Krajnya«k and Zettelmeyer (1998).8 Over time, the real exchange rate of Bulgaria has approached the level that would be in line with GDP per capita.
For Romania, the adjustment process turns out to be slightly overshooting, and from the mid-1990s onward, the real exchange rate has even become over- valued. For Russia and Ukraine, the initial undervaluations, which were consid- erably larger than for Bulgaria and Romania, were corrected much more quickly, leading to an overvaluation in 1997 in Russia, which was corrected during the crisis in 1998. Regarding Croatia and Turkey, the level exchange rates are found to be steadily overvalued.
Table 1
Cross-Sectional Regressions
Countries Year Benchmark R2
Maeso-Fernandez et al. (2004) 25 (OECD) 0.50 2002 EU-15 0.65
Coudert and Couharde (2003) 120 developing economies 0.25 2000 EU-15 0.24
Cˇiha´k and Holub (2003) 30 EU+CEE 0.90 1999 EU-15 0.90
Cˇiha´k and Holub (2003) 22 EU+CEE 0.86 2000 EU-15 0.86
Cˇiha´k and Holub (2003) 30 EU+CEE 0.94 1999 EU-15 0.89
Source: Maeso-Fernandez et al. (2004), Coudert and Couharde (2003),Cˇiha´k and Holub (2003).
Note: The coefficient is the slope coefficient from the regression. R2 stands for the goodness-of-fit of the regression. Coudert and Couharde as well as Maeso-Fernandez et al. regress the log level of the real exchange rate on the log level of relative GDP per capita, whereasCˇiha´k and Holub regress relative price levels on relative per capita GDP levels.
7 C´iha«k and Holub (2003) note that one should interpret the temporal development of data based on the International Price Comparison (IPC) program with care. The annual data are based on interpolation/extrapolation of actual price observations carried out once every three years. The error margin of such an interpolation/extrapolation may be as high as 6%. Notice also, however, that the data here are not used to derive precise misalignment figures but rather to provide some broader trends.
8 Halpern and Wyplosz (1997) and Krajnya«k and Zettelmeyer (1998) use additional variables besides productivity to investigate initial undervaluations.
Under- and Overvaluations in Terms of Productivity Levels
80 40 0
40
80
120
160
Chart 3
Source: Author,s calculations.
Note: Positive/negative values stand for overvaluation/undervaluation. CC denotes Coudert and Couharde (2003), MFOS is Maeso-Fernandez et al. (2004), and CH1, CH2 and CH3 are the three regressions taken from Cihák and Holub (2003).
CCMFOS CH1
CH2CH3 1991
Bulgaria
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002
80 40 0
40
80
120
160 1991
Croatia
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002
40 0
40
80
120
160
200 1991
Romania
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002
40 0
40
80
120
160
200 1991
Turkey
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002
0
100
200
300
400
500
600 1991
Russia
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002
0
100
200
300
400
500
600 1991
Ukraine
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002
% %
% %
% %
`´
4 Potential Sources of Real Appreciation
As shown earlier, the currencies of the countries under review are all under- valued in terms of PPP. At the same time, chart 1 and chart 4, plotting the real effective exchange rates of the countries on the basis of monthly data, reveal that the real exchange rate of some of the countries studied has, to a varying extent, undergone an appreciation during the last ten years.
Looking at the extent of the undervaluation of the level real exchange rate of different groups of goods and services for Bulgaria, Romania and Turkey may give us an idea regarding the potential sources of the real appreciation. The larg- est undervaluation can be observed for nontradable goods. The undervaluation of the real exchange rate of regulated services is considerably larger than that of market-based services. Also, goods, especially nondurable (mostly domestically produced and consumed) goods, turn out undervalued, though to a lesser extent (see chart 5). This is in line with E«gert, Halpern and MacDonald (2004), who
Log Real Effective Exchange Rates (CPI-Based), Monthly Data (1996=100)
1.5 1.2 0.9 0.6 0.3 0
0.3
0.6
Chart 4
Source: Author,s calculations based on data presented in the appendix.
1991
Bulgaria
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
0.8 0.6 0.4 0.2 0
0.2
0.4
0.6 1991
Croatia
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
0.4 0.2 0
0.2
0.4
0.6 1991
Romania
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
0.4 0.2 0
0.2
0.4
0.6 1991
Russia
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
0.2 0
0.2
0.4
0.6 1991
Ukraine
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
0.2 0
0.2
0.4
0.6 1991
Turkey
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
argue that an appreciation of the real exchange rate of transition economies has three sources: (1) the standard Balassa-Samuelson effect (market-based serv- ices), (2) the appreciation of the real exchange rate of the open sector, and (3) a trend increase of regulated prices. Such an appreciation can be viewed as an equilibrium phenomenon, as demonstrated in chart 2 by a movement from point A to point D. Of course, initial undervaluation can also explain large real exchange rate appreciation, merely reflecting adjustment to equilibrium.
4.1 The Balassa-Samuelson Hypothesis
The large undervaluation of market services reported in chart 4 may be explained by the absolute version of the Balassa-Samuelson (B-S) effect, which is generally thought to be a source of real appreciation in a successful catching- up process. According to the relative version of the B-S effect, an increase in productivity of the open sector exceeding that in the closed sector (dual produc- tivity henceforth) may go in tandem with increases in real wages in the open sector without any loss in competitiveness given that relative PPP holds in the open sector (ðEP=PÞis stable over time). Assuming wage equalization between the open and the market-based sheltered sectors, prices in the closed sector will increase. This productivity-driven inflation in market-based nontrad- ables then results in higher overall inflation and a positive inflation differential, which in turn causes the real exchange rate to appreciate.
4.1.1 Basic Assumptions: First Glance Evidence from Yearly Data
We now proceed to evaluate the extent to which the B-S effect has contributed to the appreciation of the equilibrium exchange rate from the early 1990s. The first step is to investigate whether or not the four basic assumptions which are needed for the B-S effect to hold are verified:
1. Real wages are linked to productivity in the open sector.
2. Nominal wages tend to equalize across sectors.
3. Dual productivity is linked to the relative price of market-based nontradable goods.
4. PPP holds for the open sector.
The Real Exchange Rate in Levels for Different Groups of Goods and Services in 2002
4 3 2 1
Chart 5
Source: Author,s calculations based on data drawn from NewCronos/Eurostat.
Note: NMS-10 denotes the ten new EU Member States.
NMS-10 Bulgaria Romania Turkey
Overall Goods Domestic Services
goods Market
services Real
estate Tradable
goods Regulated
services
The first three assumptions are first judged by applying ocular econometrics to annual data obtained from national accounts.9For instance, E«gert (2004) and E«gert and Halpern (2004) have shown recently that how sectors are classified into open and closed sectors does affect the results. We follow a twofold rule for separating sectors into open and closed sectors in that we consider a sector open if (1) goods in this sector are potentially subject to good arbitrage leading to price equalization across countries, and if (2) it is governed by market forces.
This yields a classification which is in contrast with MacDonald and Wo«jcik (2004) and Mihaljek and Klau (2004), who argued that tourism, trade and transportation can also be considered to belong to the open sector.10 This is the reason why we also check how sensitive the results are when classifying those sectors as open sectors.
Data available until the mid-1990s are usually based on old national accounts standards. From the mid-1990s on, national accounts data are available in the new NACE format. To cover the whole period, the NACE sectors are grouped so as to match sectors to the old standard. Exceptions are Romania and Russia.
For Romania, NACE data are available for the entire period,11while for Russia, only data based on old national accounts standards are available.12
For the old SNA classification,13three classifications for the open sector are used including (1) industry, (2) agriculture and industry, and (3) agriculture, industry, transport and telecommunications. The rest is considered as belonging to the closed sector, except for agriculture, which, if not included in the open sector, is once used as part of the open sector and once excluded because of the potential highly distorting effects of agricultural subsidies. This yields a total of six combinations of open and closed sectors (see appendix table 1).
For the new NACE classification,14the following five measures are used for the open sector: (a) manufacturing, (b) industry, (c) industry and agriculture, (d) industry, transport and telecommunications, and hotels and restaurants, and finally (e) agriculture, industry, transport and telecommunications, and hotels and restaurants. Regarding the closed sector, five alternative measures are con- sidered: (1) the remaining market-based sectors, (2) the remaining market- based sectors plus real estate, (1) and (2) augmented by agriculture if not used in the open sector, (3) market-based sectors and non-market-based sectors
9 Data are obtained from the annual database of the Vienna Institute for Comparative Economic Studies (wiiw). The database contains data broken down into five sectors for Bulgaria, Croatia, Russia and Ukraine from 1991 onward. For Bulgaria and Croatia, a 15-sector disaggregation is available from 1996, in accordance with the NACE classification. Such disaggregated data are available for Romania and Turkey for the whole period. For a detailed description of the data, see appendix 2.
10However, these sectors cannot be viewed as open sectors because, notwithstanding the relatively high share of exports, prices are determined by domestic factors in these sectors.
11It should be noted that some doubt arises regarding the reliability of such data starting in 1991.
12For Romania, data in NACE format cover the period from 1991 to 2003. For Russia, data are available only in the old format, from 1991 to 2003. Data for Bulgaria, Croatia and Ukraine are available both in the old format and in NACE format:
Bulgaria (old: 1991—96, NACE: 1996—2003); Croatia (old: 1991—95, NACE: 1995—2003); Ukraine (old: 1991—2000, NACE: 2001—03).
13The old classification provides data on six sectors: (1) agriculture, (2) industry, (3) construction, (4) transport and telecom- munications, (5) trade, (6) others.
14The NACE classification contains the following sectoral breakdown: (1) agriculture (including hunting, forestry and fishing), (2) mining and quarrying, (3) manufacturing, (4) electricity, gas and water supply, (5) construction, (6) wholesale and retail trade, (7) hotels and restaurants, (8) transport, storage and telecommunications, (9) financial intermediation, (10) real estate, renting and business activities, (11) public administration and defense and compulsory social security, (12) education, (13) health and social work, and (14) other community, social and personal services activities.
Real Wages and Productivity Growth in the Open Sector
20 10 0
10
20
30
Chart 6
Source: Author,s calculations.
Note: RW1_PPI and RW1_CPI are the PPI- and CPI-deflated nominal wages in the open sector. PROD_M and PROD_E denote average labor productivity in the open sector using data on employment (M) and on employees (E). The open sector includes industry (PROD1) or industry and agriculture (PROD2).
1991
Bulgaria
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 20 10 0
10
20
30 1991
Romania
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
20 10 0
10
20
30
40
50 1991
Croatia
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 20 10 0
10
20
30
40
50 1991
Russia
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
20 10 0
10
20
30
40
50 1991
Ukraine
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
% %
% %
%
BG_RW1_PPI BG_PROD1_M BG_PROD1_E
RO_RW1_PPI RO_PROD1_M RO_PROD1_E
CR_RW2_PPI
CR_PROD2_E RU_RW1_PPI
RU_RW1_CPI RU_PROD1_M
UKR_RW1_PPI UKR_PROD1_M UKR_PROD1_E
(education, health, public administration and other communal services) and (4) a measure of (3) completed with agriculture. This yields a total of 18 com- binations of open and closed sectors (see appendix table 2).
Growth rates of average labor productivity and real wages in the open sector are depicted in chart 6 above.15 Average labor productivity is obtained as sec- toral real value added divided by employment (PROD_E) or the number of employees (PROD_M). Real wages are calculated as the nominal wage in the open sector divided by the producer price index (PPI). As the PPI is highly distorted by oil prices in the case of Russia, the consumer price index (CPI) is used additionally for this country. Generally speaking, productivity and real wages broadly grew hand in hand, perhaps with the exception of Romania.
However, in Croatia wages rose more slowly than productivity from 2000 to 2002. In Bulgaria, Russia and Ukraine, we can observe periods during which productivity increased faster than real wages followed by periods when the opposite happened.
As far as wage equalization is concerned, the ratio of the nominal wage in the open sector to the nominal wage in the closed sector corresponding to the dual productivity differentials described above is shown in chart 7. For Bulgaria, the ratio decreased steadily over the period under study, which implies that nom- inal wages grew faster in the closed sector than in the open sector (amplification
15Note that wage data based on national accounts are not available for Turkey.
Wage Equalization across Sectors
1.4 1.0 0.6 0.2
Chart 7
Source: Author,s calculations.
1991
Bulgaria
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 1.4 1.0 0.6 0.2
1991
Croatia
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
2.2 1.8 1.4 1.0 0.6 0.2
1991
Romania
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2.2 1.8 1.4 1.0 0.6 0.2
1991
Russia
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
1.4 1.0 0.6
1991
Ukraine
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
of the B-S effect). The opposite can be observed for Russia where the ratio is on the rise (attenuation of the B-S effect). Regarding Croatia and Ukraine, jump- like changes can be observed in chart 6. Finally, the ratio is fairly stable for Romania, provided agriculture is excluded from the analysis.
4.1.2 Basic Assumptions: Econometric Evidence from Monthly Data
Using monthly data instead of annual data allows a more rigorous examination of the assumptions underlying the B-S model, which can be formulated econo- metrically as follows:
1. Productivity in the open sector is cointegrated with real wages in the open sector, with the estimated long-term coefficient being equal to 1.
2. The sectoral wage ratio is difference stationary.
3. Dual productivity is cointegrated with the relative price of market-based nontradable goods, with the estimated long-term coefficient being equal to 1.
4. The tradable price-based real exchange rate is difference stationary.
Average labor productivity is now based on industrial production and employment in industry. Real wages are obtained as gross or net monthly wages (depending on data availability) divided by the PPI (and by the CPI for Russia).
Long-term cointegration relationships are estimated using three alternative cointegration techniques: the Engle and Granger (EG) method (Engle and Granger, 1987), the Dynamic OLS (DOLS) by Stock and Watson (1993) and the error correction representation of the Autoregressive Distributed Lags (ARDL) model of Pesaran et al. (2001).16For the EG and DOLS techniques, residual-based cointegration tests are conducted, whereas the bounds-testing approach proposed by Pesaran et al. (2001) is used for the ARDL model. As an additional check to the standard cointegration tests, error correction terms are also reported for the EG and ARDL estimates. Note that we stick to this systematic assessment throughout the whole paper in order to check for the sensitivity of the results regarding different econometric specifications.
The results reported in table 2 show the existence of a long-run relationship between gross monthly real wages and productivity in the open sector for Bulgaria from 1991 to 2004. Notice that the coefficient estimates are very low, at 0.09 (EG and DOLS), and insignificant when using the ARDL approach.
The estimated coefficients are somewhat higher (about 0.45), but still consider- ably below unity for the period following the financial crisis in 1997. For Croatia and Romania, both gross and net monthly wages are available for the period from 1994 to 2004. For Croatia, cointegration can be detected unambiguously only when the bounds-testing approach is used. For both gross and net wages, the estimated long-run coefficient is slightly higher than 1. As for Romania, the relationship between productivity and real wages is fairly weak because, notwith- standing the presence of a long-term relationship, the coefficient is near zero or insignificant for gross wages. For net wages, the estimated coefficient turns out to be negative, which is in sharp contrast with our expectations. Turning to Turkey, all three estimation techniques indicate the presence of cointegration.
Except for ARDL, where the coefficient is not significant, the estimated coeffi-
16These techniques were used in e.g. E«gert (2004), where a more detailed description of the techniques can be found.
cients vary from 1.6 to 1.9. Robust cointegration for Russia can be found only when a dummy is used to capture the post-Russian crisis period from December 1998 until the end of the period, and for Ukraine when using DOLS and ARDL.
For Russia, the estimated coefficient that links productivity to real wages is pos- itive and is about 1.4. For Ukraine, the coefficient ranges from 0.7 to 1.3.
The sectoral wage ratio is defined as the ratio of nominal gross or net wages in industry to those in the whole economy. According to test results reported in table 3, the Augmented Dickey Fuller (ADF), the Phillips-Perron (PP) and the Elliott-Rothenberg-Stock (ERS) point optimal unit root tests are unable to reject the presence of a unit root, while the Kwiatowski-Phillips-Schmidt-Shin
Table 2
Cointegration Tests between Productivity and Real Wages, Monthly Data
Cointegrating vector X = [RWAGE,PROD];0= [1,1]; expected sign = [1,+]
EG DOLS ARDL EG DOLS ARDL
Gross wages 1991 to 2004 Net wages 1998 to 2004
Bulgaria
LAG (0.0), S, A, H (7.1), H (6.1), A, H (6.6), A
COINT 3.60** (7), A 3.67** (7), A, H 12.547** 3.3* (6), H, A 3.247* (0), S, A, H 4.951*
ECT 0.11*** 0.149*** 0.102* 0.144**
CONST 0.204*** 0.201*** 0.256*** 0.108** 0.121** 0.163
1 0.091*** 0.090*** 0.008 0.444*** 0.464*** 0.551*
Gross wages 1994 to 2004 Net wages 1994 to 2004
Croatia
LAG (4.0), S, H (5.0), A
COINT NO NO 9.896** NO NO 9.003**
ECT 1.36*** 0.108*** 0.107*** 0.076**
CONST 0.199*** 0.502*** 0.271*** 0.741***
1 2.064*** 1.16* 2.636*** 1.242
Gross wages 1994 to 2004 Net wages 1994 to 2004
Romania
LAG (0.0), S (3.3), S, A, H (6.0), S, H (4.0), A, H
COINT 3.756** (3), A 3.767** (3), A, H 7.861** 3.626** (3), H 3.654** (4), A 10.426**
ECT 0.139** 0.185*** 0.128** 0.208***
CONST 0.023 0.035 0.07 0.042 0.17*** 0.176*
1 0.043* 0.050* 0.023 0.038 0.128*** 0.129*
Gross wages 1993 to 2004 Russia
LAG (1.2), S, A, H (6.6)
COINT (12), 3.698** (0), 3.288* 8.213**
ECT 0.059*** 0.125***
CONST 0.33*** 0.375*** 0.388***
1 1.058*** 1.495*** 1.417***
DUMMY_98—04 0.182*** 0.134*** 0.087
Gross wages 1996 to 2004 Ukraine
LAG (5.6), S, A, H (1.0), S, A, H
COINT NO (1, all), 3.747** 5.997**
ECT 0.095*** 0.081**
CONST 0.051*** 0.103*** 0.107*
1 0.787*** 0.724*** 1.315***
DUMMY_98—04 0.176*** 0.117*** 0.362***
Gross wages 1988 to 2004 Turkey
LAG (0.4) (4.1)
COINT 3.421** (1) 3.079* (1) 4.907*
ECT 0.025* 0.028**
CONST 0.157*** 0.124*** 0.183
1 1.65*** 1.953*** 0.311
Source: Authors calculations.
Note: EG, DOLS and ARDL denote the Engle-Granger, the Dynamic OLS and the Autoregressive Distributed Lags estimations. The raw LAG shows the lag structure of the DOLS and ARDL models. S, A and H indicate that the lag structure was chosen on the basis of the Schwartz, Akaike and Hannan-Quinn information criterion, respectively. The raw COINT contains residual-based cointegration tests for the EG and the DOLS approach (with the lag length in parentheses), and test statistics from the bounds-testing approach for ARDL. The error correction terms for EG and ARDL are reported in the raw ECT.
*, ** and *** denote that the null hypothesis is rejected at the 10%, 5% and 1% levels, respectively. CONST is a constant term. NO: No cointegration could be found.
(KPSS) test mostly rejects stationarity for the whole sample and for a shorter period, i.e. 1996 to 2004, used for the sake of comparability across countries.
The only country for which there is some (mixed) evidence for difference sta- tionarity is Russia. Note also that the wage ratios based on both gross and net monthly wages exhibit trend stationarity for the subperiod. In sum, with the exception of Russia, all series either have a unit root or are trend stationary, implying the first and/or second moments to be unstable over time.
The relationship between dual productivity and the relative price of market nontradables is investigated using monthly data. Dual productivity is proxied by average labor productivity in industry or manufacturing.17For the relative price of market nontradables, three measures are employed: (1) the ratio of services in the CPI to goods in the CPI, (2) the ratio of services in the CPI to the PPI, and (3) the CPI-to-PPI ratio. Time series for services and goods in the CPI are obtained from the Main Economic Database of the OECD. As the OECD has ceased to publish these series for Bulgaria, Croatia, Romania and Ukraine, the series for these countries end at the end of 2001 or 2002.
Turning now to the estimation results shown in tables 4a and 4b, we can observe the following pattern. On the one hand, productivity and relative prices based on service prices (SERVGOODS or SERVPPI) appear cointegrated in a reasonably robust manner with coefficients of around 1 for Bulgaria18and Russia,
Table 3
Unit Root Tests for the Sectoral Wage Ratio, Monthly Data
Gross monthly wages Net monthly wages
ADF PP KPSS ERS ADF PP KPSS ERS
Bulgaria 1991:01 to
2004:03 1.13 (5) 2.23 (6) 0.39*** (10) 8.39 (5) 1996:01 to
2004:03 0.99 (4) 1.42 (6) 1.13*** (7) 10.50 (4) Croatia
1994:01 to
2004:03 1.93 (2) 2.27 (3) 0.71*** (9) 38.50 (2)
1993:01 to
2004:03 2.65 (2) 2.35 (6) 0.98*** (9) 5.84 (2) 1996:01 to
2004:03 1.69 (2) 2.14 (5) 0.39* (7) 53.30 (2)
1996:01 to
2004:03 2.06 (3) 2.52 (6) 0.44* (7) 23.15 (3) Romania
1993:01 to
2004:03 1.72 (1) 13.05 0.74** (9) 4.26 (1)
1991:04 to
2004:03 3.01** (1) 3.57** (4) 0.35* (9) 24.73 (1) 1996:01 to
2004:03 1.25 (1) 1.72 (4) 1.14*** (7) 7.13 (1)
1996:01 to
2004:03 1.26 (1) 1.93 (4) 1.08*** (7) 6.40 (1) Russia
1992:01 to
2004:03 1.35 (12) 4.88*** (7) 0.99*** (9) 13.37 (12) 1996:01 to
2004:03 5.21*** (2) 5.17*** (4) 0.64** (7) 17.83 (2) Ukraine
1996:01 to
2004:03 1.21 (1) 0.98 (3) 1.04*** (7) 79.59 (1) Source: Authors calculations.
Note: ADF, PP, KPSS and ERS are the Augmented Dickey-Fuller, the Phillips-Perron, the Kwiatowski-Phillips-Schmidt-Shin and the Elliott-Rothenberg-Stock point optimal unit root tests, respectively, for the case including only a constant. In parentheses the lag length chosen using the Schwartz information criterion is given for the ADF and ERS tests, and the Newey West kernel estimator for the PP and KPSS tests. *, ** and *** denote the rejection of the null hypothesis. For the ADF, PP and ERS tests, the null hypothesis is the presence of a unit root, whereas for the KPSS tests, the null hypothesis is stationarity.
17Productivity changes in the closed sector are set to zero because no data are available on a monthly basis.
18Overall, our results are a little more encouraging for Bulgaria than those reported in Nenovsky and Dimitrova (2002), who argued that the B-S effect did not work in Bulgaria between 1997 and 2001 because of the nonfulfillment of the underlying assumptions.
and in a less robust manner with coefficients relatively close to 1 for Croatia.
On the other hand, virtually no cointegration can be found for Romania, and the estimated coefficients are not significant for Ukraine. For Turkey, there is either no cointegration or the coefficient is fairly high, i.e. around 4. As for the
Table 4a
Cointegration Tests between Productivity and Relative Prices,
Monthly Data
Cointegrating vector X = [SERVGOODS/SERVPPI,PROD];0= [1,1]; expected sign = [1,+]
SERVGOODS SERVPPI
EG DOLS ARDL EG DOLS ARDL
1995:01 to 2002:09 1991:12 to 2002:09
Bulgaria
LAG (5.0) (1.6) (4.1)
COINT NO NO 24.247** 3.203* (1) 3.325* (1) 6.102**
ECT 0.243*** 0.209*** 0.09** 2.559 0.159***
CONST 0.12*** 0.336*** 0.104*** 0.061*** 0.289***
DUMMY_97 0.451*** 2.257*** 0.349*** 0.73*** 0.942***
1 1.155*** 0.973** 0.961*** 1.033*** 0.664**
1997:01 to 2002:09 1997:01 to 2002:09
Croatia
LAG (6.6) (6.6)
COINT 1.626 (0) NO 5.705* 3.636** (11) NO 4.515a)
ECT 0.072** 0.109** 0.063** 0.088**
CONST 0.021*** 0.015 0.084*** 0.089*
1 0.854*** 1.37** 0.845*** 1.357*
1994:01 to 2002:08 1994:01 to 2002:08
Romania
COINT 0 NO NO NO NO NO
ECT 0.005
CONST 0.582***
1 0.75***
1993:01 to 2004:03 1993:01 to 2004:03
Russia
LAG (0.0) (4.3) (0.0) (12.12)
COINT 3.523*(1) 3.431** (1) 4.994* 3.32* (0) 3.351* (0) 13.365**
ECT 0.078** 0.129*** 0.088** 0.16***
CONST 0.232*** 0.234*** 0.308*** 0.07*** 0.07*** 0.045
DUMMY_1998 0.292*** 0.288*** 0.273*** 0.469*** 0.469*** 0.442***
1 1.027*** 1.05*** 1.049*** 0.745*** 0.741*** 0.839**
1994:01 to 2004:03 1994:01 to 2004:03
Turkey
LAG (0.1) (6.2) (6.3) (6.6)
COINT 2.187 (1) 2.228 (1) 8.863** 1.114 (12) 3.257* (0) 9.422**
ECT 0.156*** 0.064 0.099** 0.101**
CONST 0.192*** 0.608** 0.121*** 0.274*** 0.357*
1 0.412 8.378* 0.964** 4.373*** 4.075
1994:01 to 2001:12 1994:12 to 2001:12
Ukraine
LAG (5.3) (0.0) (3.0)
COINT 3.501** (12) NO 7.316** 5.475** (1) 5.599** (1) 6.107**
ECT 0.06** 0.073** 0.098*** 0.084***
CONST 0.126** 0.679*** 0.107** 0.107** 0.621***
DUMMY_1998 0.466*** 0.376 0.191** 0.191** 0.357**
1 0.083 0.211 0.231 0.226 0.281
Source: Authors calculations.
Note: EG, DOLS and ARDL denote the Engle-Granger, the Dynamic OLS and the Autoregressive Distributed Lags estimations. The raw LAG shows the lag structure of the DOLS and ARDL models. The raw COINT contains residual-based cointegration tests for the EG and the DOLS approach (with the lag length in parentheses), and test statistics from the bounds-testing approach for ARDL. The error correction terms for EG and ARDL are reported in the raw ECT. *, ** and *** denote that the null hypothesis is rejected at the10%, 5% and 1% levels, respectively. CONST is a constant term. a) indicates ambiguity in the sense that the tests statistic lies in a range where there is no clear indication of the absence or existence of a cointegrating relationship (Pesaran et. al., 2001). NO: No cointegration could be found.
CPI-to-PPI ratio (CPIPPI), cointegration with the good sign could be estab- lished not only for Bulgaria, but also for Croatia, Turkey and Ukraine, albeit with fairly low coefficients in the latter country.
Finally, unit root tests including a constant are reported in table 5, from which it can be seen that the PPI-based real exchange rate is clearly not dif- ference stationary in levels for Bulgaria, Croatia, Romania and Ukraine. For Russia, the null of a unit root cannot be rejected by the ADF, PP and ERS tests, and the KPSS test is not able to reject the null of stationarity. The opposite happens to be the case for Turkey, where the ADF, PP and ERS tests suggest difference stationarity. However, the KPSS test indicates nonstationarity.
Thus, it is fair to say that PPP does not hold for the open sector for most of the countries.
Table 4b
Cointegration Tests between Productivity and Relative Prices, Monthly Data
Cointegrating vector X = [CPIPPI,PROD];0= [1,1]; expected sign = [1,+]
EG DOLS ARDL EG DOLS ARDL
Bulgaria Russia
1991:12 to 2004:03 1993:01 to 2004:03
LAG (0.3) (1.4) LAG (0.0) (2.4)
COINT 3.734** (1) 3.217* (1) 12.978** COINT 4.25** (0) 5.028** (2) 7.895**
ECT 0.063*** 0.073*** ECT 0.202*** 0.263***
CONST 0.187*** 0.206*** 0.032 CONST 0.117*** 0.118*** 0.132***
DUMMY_97/98 0.208*** 0.197** 0.899*** DUMMY_97/98 0.152*** 0.154*** 0.17***
1 0.49*** 0.514*** 0.068 1 0.343*** 0.353*** 0.406***
Croatia 1992:01 to 2004:03
Ukraine 1994:12 to 2004:03
LAG (6.6) (6.0) LAG (0.0) (6.0)
COINT 3.441** (10) 3.886** (0) 50.524** COINT 3.795*(11) 3.836** (11) 12.29**
ECT 0.098** 0.061* ECT 0.044 0.07**
CONST 0.013*** 0.012*** 0.063*** CONST 0.037*** 0.037*** 0.18***
DUMMY_98 DUMMY_98 0.029 0.029 0.152**
1 0.679*** 0.716*** 0.445 1 0.132*** 0.133*** 0.196
Romania 1994:01 to 2004:03
Turkey
1985:03 to 2004:03
LAG (10.12) LAG (4.0) (6.6)
COINT 2.408 (0) NO 13.11** COINT 4.673** (1) 4.267** (0) 5.084*
ECT 0.062** 0.006 ECT 0.131*** 11 0.111***
CONST 0.029 3.303 CONST 0.028*** 0.004 0.037
DUMMY DUMMY 0.145*** 0.118*** 0.101*
1 0.065*** 1.354 1 0.569*** 0.676*** 0.673**
Source: Authors calculations.
Note: As for table 4a.
Table 5
Unit Root Tests for the PPI-Based Real Exchange Rates, Monthly Data
ADF PP KPSS ERS
Bulgaria 1993:01 to 2004:03 2.084 (0) 1.992 (2) 0.979*** (9) 3.104* (3) Croatia 1993:01 to 2004:03 1.337 (1) 1.290 (3) 0.764*** (9) 7.719 (1) Romania 1994:01 to 2004:03 1.686 (0) 1.592 (6) 1.025*** (9) 15.797 (0) Russia 1994:01 to 2004:03 1.854 (1) 2.078 (6) 0.169 (9) 11.840 (1) Turkey 1985:01 to 2004:03 3.138** (0) 3.376** (2) 0.412* (11) 3.750* (0) Ukraine 1996:01 to 2004:03 1.088 (2) 1.052 (2) 0.845** (7) 20.567 (2) Source: Authors calculations.
Note: ADF, PP, KPSS and ERS are the Augmented Dickey-Fuller, the Phillips-Perron, the Kwiatowski-Phillips-Schmidt-Shin and the Elliott-Rothenberg-Stock point optimal unit root tests, respectively, for the case including only a constant. In parentheses the lag length chosen using the Schwartz information criterion is given for the ADF and ERS tests, and the Newey West kernel estimator for the PP and KPSS tests. *, ** and *** denote the rejection of the null hypothesis at the 10%, 5% and 1% levels. For the ADF, PP and ERS tests, the null hypothesis is the presence of a unit root, whereas for the KPSS tests, the null hypothesis is stationarity.
4.1.3 To Sum Up
All in all, there is mixed evidence regarding the functioning of the basic assump- tions. First, increases in productivity are connected to increases in real wages in the open sector roughly proportionately only in Croatia, Russia and Ukraine.
The effect of productivity on real wages is below 1 in Bulgaria, and the relation- ship is highly questionable for Romania. By contrast, changes in productivity in the open sector lead to disproportionately large changes in real wages in Turkey.
Second, a proportionate wage equalization between the open and closed sectors can be verified to a limited extent only for Russia. Third, notwithstanding the mixed evidence on real wages and nominal wage equalization, the service-based relative price is found to be linked reasonably well to dual productivity with a coefficient in the neighborhood of 1 for Bulgaria, Croatia and Russia. The coef- ficient is much higher than 1 for Turkey and considerably lower than 1 for Ukraine. No cointegration could be detected for Romania. Overall, this sug- gests that the B-S effect works reasonably well in Bulgaria, Croatia and Russia, whereas it is attenuated in Ukraine and is amplified in Turkey. For Romania, it does not seem to function. Another question is, however, the influence of the B-S effect on overall inflation, an issue which is addressed in the next section.
Fourth, relative PPP is rejected for the real exchange rate of the open sector for all economies, perhaps with the exception of Turkey, which implies that the B-S effect will not be able to explain the entirety of real exchange rate movements.19
4.1.4 A Simple Accounting Framework
We now set out to analyze the size of the inflation to be attributed to the B-S effect (PBS). For this purpose, let us consider the following equation used in E«gert (2004):
PBS ¼ ð1Þ1ðPRODTPRODNTÞ ð1Þ
where (1) is the share of nontradables in the consumer basket, 1 concep- tually corresponds to the estimated coefficient from tables 4a and 4b, which connects the relative price of nontradables to productivity, and which, ideally, should be 1. PROD is the average labor productivity in the tradable (T) and nontradable (NT) sectors.
Average annual growth rates of the different measures of dual productivity are computed for the countries under consideration using annual data from national accounts for two periods, 1991—2001/2003 and 1996—2001/2003.
For Turkey, the series start in 1970. This is why two additional periods are considered for this country, namely 1970—2003 and 1970—90.20 In addition, average annual growth rates are computed using monthly industrial produc- tion-based productivity measures.21
The results are displayed in tables 6a to 6d. Several observations deserve attention. The first observation is that it may matter whether average labor pro-
19If relative PPP were verified for the open sector, then the B-S effect could explain real exchange rate movements based on the CPI.
By contrast, if relative PPP cannot be verified, the B-S effect will provide an explanation for changes in the difference between the (CPI-based) overall real exchange rate and the real exchange rate of the open sector.
20It should be mentioned that the productivity figures may be biased downward for Russia and Ukraine because from 1995 to 1998, huge numbers of employees were forced to take unpaid leaves. Hence, they are included in the statistics even if they did not contribute to output.
21The same periods were considered here as for the national accounts-based data. For Croatia, Romania and Russia, data for 2003 (not available from national accounts) are also shown for comparsion purposes.
